Integrating the planning of sporadic and slow-moving parts within the normal planning process

Transcription

1 Demand Planning Integrating the planning of sporadic and slow-moving parts within the normal planning process Managing the forecasts for sporadic and/or slow-moving parts within normal business forecasting processes can be a challenge. Techniques have been developed to forecast demand and manage the inventory for such items. These techniques have proven to be generally applicable for many items with a sparse demand pattern not just true slow movers, but also those items that sell in large volumes when they do sell.

2 Table of Contents 3 Business requirements 6 The solution 7 Coping with the certainty of SMP forecasts 8 Key benefits

3 Business requirements Existing time series forecasting techniques have proven to be invaluable to many companies that supply the market from stock, rather than by order. Such techniques have evolved with the advent of computing power from simple exponential techniques to modern applications that use Bayesian techniques and apply dynamic linear models (DLM) to the demand history to produce forecasts. 1 The typical requirements of a business application can be characterized as follows: Need to be able to use the normal business periodicity to report the sales/demand history to produce forecasts. Need to be able to produce a forecast by period that fits with this business reporting periodicity. Need the forecast to be general purpose. It should drive the planning processes and be used in forward estimates of the business by item or group. It may also be the budget or form the basis of the budget and targets The forecast should also be capable of driving an inventory policy that is based around the actual inaccuracies seen between the past forecast and actual sales/demand. All this takes place against a backdrop of continual new product introductions and supercession. A traditional time series approach provides a good solution to the requirements listed above for items that are regular movers, but it falls short when dealing with sporadic or slow-moving parts (SMPs). A typical SMP may be an item at a location that sells one or two units in a period and then there will be a number of periods with no sales, until the next sales event. Some businesses create many SMPs in their planning process by forecasting at locations close to the customers, rather than simply producing forecasts at stocking locations. Recent trends in vendor-managed inventory (VMI) and collaborative forecasts have accentuated this issue for many forecast analysts. Other companies produce SMPs by the minimum lot quantities that they impose on their customers. Such SMPs could be dealt with by structural changes in the forecasting processes or by changing the ordering lot quantities. History chart Year Jan Feb Mar Apr May Jun July Aug Sept Oct Nov Dec Figure 1: Sporadic or slow moving part history. 3

4 However, the business benefits of VMI processes and collaborative forecasting mean that the market levels that the forecasting process is run at will become increasingly focused toward the consumer level. Even if a company can solve all these issues, there will still be some SMPs left that require forecasting. This could range from 10 to 50 SKUs in a pharmaceutical business to hundred of thousands for a service parts application. At the top end of this scale, it is obvious that any SMP process should give robust and reliable forecasts that can be produced with minimum intervention. It should also be possible to use such forecasts for procurement and production planning, using techniques from a simple ROP process through to full-blown MPS planning. Often, the only practical approach when dealing with SMPs using traditional time series analysis is to identify a level and perhaps trend, but with no seasonality. This produces a simple forecast that, at best, gives an estimate of average sales per reporting period. What it fails to indicate is how big those sales may be in any period where an order is received. A further problem with this approach is that the measure of error (standard deviation) between the forecast and the actual demand is very poor, as it is strongly influenced by the many periods where no demand was seen. This dilutes the measure of variance that the business should be interested in the comparison between the forecast of the sales expected in the next period with some demand and the demand actually seen. The other main problem with this approach is the high level of manual effort required to produce accurate, robust, and useable business forecasts. Table of history and forecasts Jan Feb Mar Apr May Jun July Aug Sep Oct Nov Dec Total Outlook 209 Figure 2: Traditional forecasting technique applied to slow moving product. 4

5 Figure 2 shows a simple regression model with no dynamic smoothing applied through a sparse history set in figures, July 2012 being the first period to forecast. This approach only gives an estimate of approximately 10 units per month, with a steady (and perhaps spurious) growth; there is no indication of how many units may be sold when the next sales event occurs. Also the technique does not take into account the obvious step in the history data; a more realistic average monthly forecast would be 7 or 8 units per month. One last point to consider is that the standard deviation produced for this item is approximately 12 units, which will have a major effect on the statistically derived safety stock. Businesses require the ability to use a forecasting system that can identify normal and slow movers automatically and apply forecasting techniques tailored to the type of item being forecasted. Ideally, such a system should employ modern and powerful time series techniques to forecast the normal items. The system should then be able to switch to SMP techniques automatically, utilizing the same periodicity of history as supplied for the normal items and with a minimum requirement for other information, control fields or user intervention. The SMP techniques should be targeted at providing an estimate of the demand the business should expect in the next period in which there is demand. It should also be targeted at producing a good statistical measure of variance between the expected demand in any period with sales and the actual demand seen. In addition to this, the technique should provide the information required to drive the other business processes that require a forecast, such as target and budget setting and group reporting. Also, the process should require minimum analyst intervention and be very fast at processing each item. These requirements are obviously true of service parts suppliers, but our work with other industries has convinced us that most, if not all, companies supplying the market from stock have some SMPs and that the planning of these items would benefit from better techniques. 5

6 The solution Various solutions have been proposed in the past to deal with SMPs. The process that we have developed was evolved in house from first principles, but is closely related to a solution proposed by Croston, and expanded by Johnston and Boylan. 2,3 Croston suggests that each event should be monitored and that an exponentially weighted smoothed average can be generated through the quantities seen in each event. This average then gives an estimate of the next event that will be seen. He also proposes that a smoothed average should be created to estimate the gap between events. One of the key differences in our approach is the assumption that demand in any business reporting period is a logical event. This approximation allows the use of the normal history reporting mechanism in a business, and allows businesses to work from one set of history data, so they can join the normal and SMP forecasting processes in one system. The other key difference is the forecast process applied. When an SMP is detected, two forecasts are generated. The first forecast is an estimate of the average event size and the second forecast is an estimate of the gap between the logical events. Both of these models are single term DLMs. This forecasting approach allows the development of dynamic models that react to the most recent events seen. The sensitivity of this reaction is under user control, but default discount factors (smoothing ratios) are applied that make the models highly dynamic. The estimate of the next logical event is the forecast that is used to produce reorder point (ROP) quantities. The result of multiplying the estimate of the next event and the chance of an event gives a good average period forecast for the other business processes that need weekly or monthly estimates. The ROP uses a different calculation for SMP items than for normal ones. The ROP for a standard product is typically the forecast over the re-supply lead-time, plus the required safety stock. When applying the SMP process, the ROP calculation needs adjusting to be the number of potential events in the lead-time multiplied by the forecasted average event quantity, plus the required safety stock. In this example, the chance of an event during a month is about 0.3, or just less than 1 event every three months. Figure 3 shows the history and average monthly forecasts for the item. This average monthly projection can be used by the normal business forecasting processes, such as budget creation or the reporting of aggregated forecasts across groups of items. 6

7 Table of history and forecasts Jan Feb Mar Apr May Jun July Aug Sep Oct Nov Dec Total Outlook 94 Figure 3: SMP forecasting technique. Coping with the certainty of SMP forecasts While the SMP forecast gives us a measure of both the size of an event and the chance of it occurring during a lead time, it does not account for the uncertainty inherent in the forecast. In the previous section the concept of reorder point (ROP) was introduced and defined, and it is the safety stock element of the ROP that provides protection when the actual demand is in excess of the forecast. Traditionally, this safety stock has been set at a number of months cover, or at a predefined level that is usually based on a convenient quantity to hold a case, for example. Neither of these strategies properly addresses the issue of balancing inventory investment with customer service levels, both key performance indicators. The key to setting safety stock is to measure the uncertainty in the forecast, and to take account of inbound replenishment frequency and lead time. Given these factors, a stock level can be calculated to maximize customer service from a given investment in inventory. The general effect of this strategy is to rebalance inventory so that relatively stable, short lead time items have less stock, and volatile, long lead time items have more stock. Figure 4 illustrates this effect. With the ability to calculate both SMP forecasts and variability quickly and with little user intervention, it is now possible to use sophisticated safety stock setting rules to effectively manage a large inventory. 7

8 Key benefits Using the same technique to forecast both normal items and SMPs can be problematic, and require extensive analyst intervention. Methods specifically designed for SMPs can provide a good result, but have the disadvantage of working at a level outside that of a company s normal forecasting cycle. The proposed methodology allows the automatic treatment of SMPs as a special case of the DLM, within the existing reporting periodicity. We believe the benefits of this approach include: Reduced analyst effort in managing sometimes very large numbers of SMPs More realistic forecasts for SMPs, especially when used for annual and group totals Lower forecast error, resulting in better safety stock calculations ROP takes into account both size and chance of an event within the lead time Replenishment plans can be easily produced from the results of the SMP analysis Consistent approach across the whole inventory These improvements bring real value to the business, including: Reduced inventory holding for SMPs Improved customer service Fewer expediting costs Faster lead time to the consumer Reduced obsolescence Rebalancing inventory Investment Original New Figure 4: The effects of rebalancing inventory. Weeks of Stock Category 8

9 Fore more information, visit Infor Supply Chain Management 1 Andy Pole, Mike West and Jeff Harrison, Applied Bayesian Forecasting and Time Series Analysis, Chapman & Hall, J.D. Croston, Forecasting and Stock Control for Intermittent Demand, Operational Research Quarterly, Vol. 23, 1972, pp F. R. Johnston and J. E. Boylan, Forecasting for Items with Intermittent Demand, Journal of the Operational Research Society. Vol. 47, 1996, pp Share this : Copyright 2016 Infor. All rights reserved. The word and design marks set forth herein are trademarks and/or registered trademarks of Infor and/or related affiliates and subsidiaries. All other trademarks listed herein are the property of their respective owners Avenue of the Americas, New York, NY INFDTP en-US